Colorization transformer: A hybrid deep learning approach for grayscale image colorization

Gray scale image colorization is key challenging computer vision, with applications in digital restoration, content creation, and historical documentation. While CNN shave advanced the field, they often struggle with semantic consistency and global coherence in complex scenes. We propose a hybrid deep learning model that combines a CNN encoder for local feature extraction with a Transformer encoder for modeling long-range dependencies. A CNN decoder then reconstructs high-resolution color images from the learned representations. The model is trained on a dataset of 7000 grayscale-RGB image pairs, normalized for robustness. We optimize using a combination of L1 loss and Structural Similarity Index (SSIM)to ensure both perceptual accuracy and structural fidelity. Our approach out performs traditional methods on key metrics such as SSIM and PSNR, generalizing well to unseen inputs while producing vibrant and semantically accurate colorizations suitable for real-world applications.